Key words: Interpolation, Differential GPS, Digital Photogrammetry, Accuracy, Geomorphology
Digital elevation models (DEMs) are key components of computer-based analyses of drainage basins, providing elevation information for the land surface throughout the catchment. Numerous methods exist for producing DEMs including interpolating from contour data, interpolating from differential Global Positioning System (dGPS) point data and Digital Photogrammetry techniques. Data on the accuracy of these three methods tends to be scarce and sometimes conflicting. This research aims to identify some of the factors involved and quantitatively compare the accuracy of the different techniques.
Paper maps have traditionally used contour data to represent the Earth's surface, but converting contour data to a DEM surface is not a simple process. Many techniques have been developed to interpolate DEM data from other sources and a number are available with the GIS software examined here, e.g., ArcView and Idrisi. Little is known about the quality of the output from these techniques. A variety of interpolation algorithms are quantitatively compared, including inverse distance weighting, kriging, and linear interpolation.
Differential GPS data are a relatively new approach to surveying which has a very high reported accuracy from the suppliers. Independent research has shown potential errors an order of a magnitude larger than reported. The 'high tech' nature of dGPS has resulted in many operators taking a 'black box' approach where they inherently believe their data is correct because 'the dGPS says so.' The approach of interpolating from dGPS point data to produce a DEM is examined and assessed.
Digital photogrammetry on computer-based digital aerial photography has been possible for a number of years now, but the advent of cheap and very powerful desktop PCs has brought the technique out of the research arena and into everyday use. The technique is now widely available, but many users may be unaware of the potential inaccuracies in the approach.
This work quantitatively compares all these techniques for producing DEMs from a geomorphological perspective, using a statistical approach similar to the Lowess best-fit line technique. This technique uses a locally weighted average as the basis for measuring local variance along a digitised vector line data set. These variance values are then averaged over a geomorphological feature such as a river long-profile, a river valley cross-section, a valley flood plain, or badlands cross section. The resulting value gives a relative statistical measure of error for each DEM, allowing quantitative comparisons of each approach and interpolation technique.
Data used in this approach are from a number of sources for regions with different climates and substrates. These include the concave-convex topography of chalklands in Hertfordshire, an actively eroding river system in the Welsh uplands, and badlands topography on poorly consolidated sediments in a semi-arid region of Spain. This approach will enable a quantitative estimate of error caused by varying surface and vegetation conditions to be made for each technique in each area.
The results should demonstrate the most accurate technique for DEM production in any of the terrain types covered and indicate the error levels associated with such techniques.